Hydroelectric equipment fault analysis method based on industrial internet platform
By combining working conditions, trends and correlation analysis on the industrial Internet platform, the data processing capabilities and depth of traditional fault diagnosis and analysis methods are solved, and the in-depth mining and accurate diagnosis of hydropower equipment faults is realized, and the fault analysis is improved. Efficiency and accuracy provide a scientific basis for the operation and maintenance management of hydropower stations.
Patent Information
- Application Number
- CN202410839226.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-26
- Publication Date
- 2024-11-22
AI Technical Summary
Traditional hydropower equipment fault diagnosis and analysis methods are limited by manpower and resources, and it is difficult to process large-scale data, resulting in insufficient comprehensiveness and reliability of the analysis results, and cannot deeply reveal the complex relationships and patterns behind the data, making it difficult to meet all-round and multi-level faults. diagnostic requirements.
The fault analysis method based on the industrial Internet platform is adopted, including working condition analysis, time trend analysis, frequency trend analysis and correlation analysis, combined with adaptive weight time series modeling, spectral clustering and mutual information calculation, to generate comprehensive reports and visual charts , dig deep into the patterns and associations behind data.
Improve the depth and efficiency of fault analysis, realize accurate diagnosis and prediction of hydropower equipment faults, enhance the timeliness and accuracy of operation and maintenance decisions, reduce unplanned downtime, and improve the operation stability and economicality of the power station. .
Smart Images

Figure CN119004301A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of hydropower equipment failure analysis, and in particular to a hydropower equipment failure analysis method based on an industrial Internet platform. Background Art
[0002] In the field of operation and maintenance of hydropower stations, fault diagnosis and analysis is a crucial task, which is directly related to the safe and stable operation and economic benefits of the power station. Traditionally, this process relies on the historical data automatically recorded by the online monitoring system and the monitoring system of the hydropower station. These data cover multiple key indicators such as the vibration, swing, pressure pulsation, operating parameters (such as head, active power), temperature, gas content in oil, etc. of the unit.
[0003] At present, data analysis mainly adopts manual statistical methods. Specifically, technicians will select parameters related to specific fault analysis from huge historical data sets, such as vibration amplitude, bearing temperature, etc., and then summarize these data and generate charts. Although this process can provide a preliminary visual display, it is often limited by human and technical means and can only perform relatively basic comparative analysis.
[0004] However, traditional manual statistical analysis methods face many challenges: due to manpower and resource limitations, the amount of data that can be included in the analysis is usually small, which directly affects the comprehensiveness and reliability of the analysis results. The time span of the analysis is often limited, making it difficult to capture long-term trends and cyclical changes, which is particularly unfavorable for predicting potential failures. The currently used analysis algorithms are relatively basic and lack advanced data mining and machine learning techniques, making it difficult to reveal the complex associations and patterns hidden behind massive amounts of data. Although traditional chart presentation methods are intuitive, they are unable to cope with multi-dimensional, high-density data and are difficult to effectively convey the deep relationships between data. Limited by the above factors, existing methods have obvious shortcomings in the depth and breadth of fault analysis, and it is difficult to meet the comprehensive and multi-level diagnostic needs of faults.
[0005] In view of the above problems, there is an urgent need for a more advanced and efficient data analysis method that can make full use of big data and artificial intelligence technology to achieve in-depth mining and accurate diagnosis of hydropower equipment failures. This requires not only improving data processing capabilities, broadening the time span and data volume of analysis, but also introducing complex analysis algorithms and innovative visualization methods in order to achieve a comprehensive analysis of the causes of failures and predictive failure prevention. Summary of the invention
[0006] The main purpose of the present invention is to provide a hydropower equipment fault analysis method based on an industrial Internet platform to solve the problem that traditional fault diagnosis and analysis methods have been difficult to adapt to the high requirements of modern hydropower station operation and maintenance management.
[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is: a hydropower equipment fault analysis method based on the industrial Internet platform, the analysis method steps include working condition analysis, time trend analysis, frequency trend analysis and correlation analysis;
[0008] Working condition analysis: based on the characteristic value of the measuring point and the historical data of the working condition, determine the interval between the abnormal data of the measuring point and the working condition;
[0009] Time trend analysis: according to the characteristic value of the exceeding measurement point and the historical data of the corresponding working conditions, the characteristic value of the exceeding measurement point within the time length is taken according to different time lengths, and a time-characteristic value trend chart is generated to analyze the development trend of the characteristic value of the measurement point;
[0010] Frequency trend analysis, based on the characteristic value of the measuring point and the data of a certain time length of the corresponding working condition, divides the characteristic value within the time into continuous intervals, counts the frequency of the characteristic value of the measuring point in different characteristic value intervals under all working conditions, generates an amplitude-frequency histogram, and determines the frequency of the characteristic value in a certain amplitude interval.
[0011] Correlation analysis: perform correlation analysis on the eigenvalue time series data of different measuring points to generate a correlation matrix diagram of the eigenvalues of different measuring points. Based on the correlation results, determine whether there is a correlation between the measuring points and estimate which factor is related to the cause of the exceeding of the standard.
[0012] In the preferred solution, the specific steps of the working condition analysis include:
[0013] A1. Classify the characteristic values of the measuring points and the corresponding working condition information of the measuring points according to the time length, generate a bubble chart of active power-measuring point characteristic value-time period, set the bubble color according to the time length, and analyze and determine the active power and time interval when the measuring point is abnormal;
[0014] A2. Classify the operating point and the characteristic value of the measuring point at the operating point, generate a bubble chart of active power-water head-measurement point characteristic value, set the bubble color according to the characteristic value, and analyze and determine the interval of water head and active power when the measuring point is abnormal;
[0015] A3. For the points where the characteristic value data of the measuring points exceed the standard, the points are segmented according to their characteristic values, and a bubble chart of active power-water head-measuring point characteristic value is generated to further determine the operating condition range where the data exceeds the standard.
[0016] In the preferred solution, steps A1-A3 in the working condition analysis are comprehensively analyzed, and the analysis results are fed back to the adaptive weight time series modeling. The weight is dynamically adjusted according to the correlation strength between the characteristic value and the working condition. The final adjusted weight is:
[0017] w i ″=αw i′+(1-α)(β·TimeImpact i +(1-β)·ConditionImpact i ) (1);
[0018] Among them, w i " is the final adjusted weight, w′ is the weight adjusted based on the degree of abnormality, α is the abnormality weight coefficient, β is the time impact coefficient, TimeImpact i and ConditionImpact i They are the degree of influence based on the analysis of time and operating point;
[0019] After analyzing the impact, a comprehensive report is generated, including time series analysis charts, operating point characteristic charts, characteristic value segmentation statistics charts, and comparison charts before and after weight adjustment;
[0020] Adjust weights based on the generated comprehensive report, retrain the time series model using the adjusted weights, and evaluate the model's performance through cross-validation to ensure that the model can more accurately reflect the actual working conditions.
[0021] In the preferred solution, spectral clustering is used to analyze the time trend. The specific steps are:
[0022] B1. Construct the time series matrix T. The time series matrix T is expressed as:
[0023]
[0024] Among them, t ij represents the characteristic value at the i-th time point and the j-th operating condition, m represents the number of time points, and n represents the number of operating conditions;
[0025] B2. Normalize the time series matrix T. Spectral clustering also includes constructing a similarity matrix W, calculating a degree matrix D, and a Laplace matrix L;
[0026] The similarity matrix W reflects the similarity between nodes;
[0027] The diagonal elements D ii is equal to the sum of all edge weights of the i-th node, where the diagonal element D ii for:
[0028]
[0029] The standardized matrix is clustered using the spectral clustering algorithm to identify trend patterns, and the Plath matrix L is calculated from the degree matrix and the similarity matrix: L = DW;
[0030] The goal of spectral clustering is to minimize the function y on the Laplacian matrix L T Ly, where y is an indicator vector used to indicate which cluster each node belongs to. The optimal clustering result y is expressed as:
[0031] y = argmin y y T Ly(4);
[0032] Among them, y is the optimal clustering result;
[0033] B3. Solve for the eigenvector of L and select the eigenvectors corresponding to the first k smallest non-zero eigenvalues for clustering, where k is the expected number of clusters.
[0034] In the preferred solution, in the frequency trend analysis, the sliding window technology is used to calculate the frequency of the characteristic value interval. The specific method is:
[0035] C1. Define the sliding window size and step size, slide the window on the time series, and calculate the distribution of feature values in each window;
[0036] C2. Use the chi-square test to assess distribution differences and identify abnormal frequencies.
[0037] In the preferred solution, in the correlation analysis, mutual information is used to calculate the correlation between the characteristic values of different measuring points. The specific method is:
[0038] D1, collects characteristic value time series data from each measurement point for a period of time to calculate probability distribution and mutual information;
[0039] D2. For any two measuring points X and Y, calculate the joint probability distribution of the eigenvalues, discretize the data, map the continuous eigenvalues into a finite interval, then count the number of samples falling on each interval combination, and finally normalize to obtain the joint probability distribution;
[0040] D3. Calculate the marginal probability distributions P(X) and P(Y). The marginal probability distribution reflects the probability distribution of the eigenvalue of a single measuring point, that is, the probability of the eigenvalue of a certain measuring point occurring without considering other measuring points. It is obtained by summing the joint probability distribution along another dimension.
[0041] D4. Apply the mutual information formula to calculate the correlation. Use the mutual information formula to quantify the correlation between the eigenvalues of the measurement points X and Y. The mutual information formula is:
[0042]
[0043] Where I(X;Y) represents the mutual information of X and Y, P(x,y) is the joint probability, and P(x) and P(y) are the marginal probabilities respectively;
[0044] D5, repeat the above D1-D4 process for all measurement points in pairs to obtain a complete mutual information matrix;
[0045] Visualize this matrix as a heat map or network diagram to intuitively show the strength of correlation between the eigenvalues of the measurement points;
[0046] D6. Based on the mutual information matrix, identify which measurement point eigenvalues are highly correlated and which measurement points are relatively independent; high mutual information values indicate direct or indirect causal relationships, while low mutual information values indicate that the measurement point eigenvalues are almost independent.
[0047] The present invention provides a hydropower equipment fault analysis method based on the industrial Internet platform. The present invention proposes an innovative fault analysis method, which integrates the powerful power of the industrial Internet platform to achieve in-depth analysis and precise positioning of hydropower equipment faults. Its core advantages and beneficial effects include:
[0048] Data integration and analysis speed-up: The Industrial Internet platform can seamlessly access real-time and historical data of unit status, greatly enriching the data source and improving the diversity and integrity of data. The efficient data processing mechanism ensures the acceleration of the analysis process, greatly shortens the fault detection and response time, and enhances the timeliness of operation and maintenance decisions.
[0049] Intelligent analysis and deep insight: The introduction of advanced working condition, trend and correlation analysis algorithms, combined with big data analysis technology, can deeply explore the complex patterns and potential associations hidden behind the data and enhance the depth of fault analysis. Through machine learning models to identify fault characteristics, automatic classification and clustering, it provides strong support for the precise location of the root cause of the fault.
[0050] Visualization enhances understanding: Innovative visualization tools and interface design transform abstract analysis results into intuitive charts and reports, helping technicians quickly understand and interpret complex data relationships, improving communication efficiency and decision-making quality.
[0051] Fault prediction and prevention: Based on trend analysis of historical data, this method can predict the types of faults that may occur in the future and their probabilities, providing a scientific basis for the formulation of preventive maintenance strategies, reducing unplanned downtime, and ensuring the stable operation of the power plant.
[0052] Decision support and optimization: Through a comprehensive analysis of the causes of failures, the operation and maintenance team can be provided with specific suggestions for improvement measures, optimize equipment maintenance plans, reduce operating costs, and improve the overall operating efficiency and safety of the power station.
[0053] Improve the breadth and flexibility of analysis: The method covers multi-level fault analysis from a single component to the entire system. It is suitable for both rapid troubleshooting of short-term anomalies and in-depth research on long-term performance degradation, meeting the fault diagnosis needs in different scenarios.
[0054] The hydropower equipment special fault analysis method based on the industrial Internet platform proposed in the present invention has significantly improved the efficiency, accuracy and intelligence level of fault analysis through data-driven technological innovation, injected new vitality into the operation and maintenance management of hydropower stations, and effectively promoted the industry to develop in a more intelligent direction. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:
[0056] Figure 1 is a flow chart of the fault analysis method of the present invention;
[0057] Figure 2 It is a flow chart of the working condition analysis method of the present invention. DETAILED DESCRIPTION
[0058] Example 1
[0059] like Figure 1-2 As shown, a hydropower equipment fault analysis method based on the industrial Internet platform, the analysis method steps include working condition analysis, time trend analysis, frequency trend analysis and correlation analysis;
[0060] Working condition analysis: based on the characteristic value of the measuring point and the historical data of the working condition, determine the interval between the abnormal data of the measuring point and the working condition;
[0061] Time trend analysis: according to the characteristic value of the exceeding measurement point and the historical data of the corresponding working conditions, the characteristic value of the exceeding measurement point within the time length is taken according to different time lengths, and a time-characteristic value trend chart is generated to analyze the development trend of the characteristic value of the measurement point;
[0062] Frequency trend analysis, based on the characteristic value of the measuring point and the data of a certain time length of the corresponding working condition, divides the characteristic value within the time into continuous intervals, counts the frequency of the characteristic value of the measuring point in different characteristic value intervals under all working conditions, generates an amplitude-frequency histogram, and determines the frequency of the characteristic value in a certain amplitude interval.
[0063] Correlation analysis: perform correlation analysis on the eigenvalue time series data of different measuring points to generate a correlation matrix diagram of the eigenvalues of different measuring points. Based on the correlation results, determine whether there is a correlation between the measuring points and estimate which factor is related to the cause of the exceeding of the standard.
[0064] In the preferred solution, the specific steps of the working condition analysis include:
[0065] A1. Classify the characteristic values of the measuring points and the corresponding working condition information of the measuring points according to the time length, generate a bubble chart of active power-measuring point characteristic value-time period, set the bubble color according to the time length, and analyze and determine the active power and time interval when the measuring point is abnormal;
[0066] A2. Classify the operating point and the characteristic value of the measuring point at the operating point, generate a bubble chart of active power-water head-measurement point characteristic value, set the bubble color according to the characteristic value, and analyze and determine the interval of water head and active power when the measuring point is abnormal;
[0067] A3. For the points where the characteristic value data of the measuring points exceed the standard, the points are segmented according to their characteristic values, and a bubble chart of active power-water head-measuring point characteristic value is generated to further determine the operating condition range where the data exceeds the standard.
[0068] This method uses historical data obtained from the industrial Internet platform, combined with operating condition analysis, time trend analysis, frequency trend analysis and correlation analysis, to achieve in-depth thematic analysis of unit failures, effectively improving the efficiency, accuracy and comprehensiveness of fault analysis, and providing strong data support for quickly locating the fault operating condition interval and exploring the cause of the fault.
[0069] Using the above methods, analysts can more efficiently identify and understand the patterns and trends of unit status changes, provide a scientific basis for fault prevention, maintenance planning and system optimization, and thus improve the operational stability and economy of hydropower equipment.
[0070] Example 2
[0071] Further illustrate with reference to Example 1, Figure 1-2 In the structure shown in the figure, steps A1-A3 in the working condition analysis are comprehensively analyzed, and the analysis results are fed back to the adaptive weight time series modeling. The weight is dynamically adjusted according to the correlation strength between the eigenvalue and the working condition. The final adjusted weight is:
[0072] w i ″=αw i ′+(1-α)(β·TimeImpact i +(1-β)·ConditionImpact i ) (1);
[0073] Among them, w i ″ is the final adjusted weight, w′ is the weight adjusted based on the abnormality level, α is the abnormality weight coefficient, β is the time impact coefficient, and TimeImpact i and ConditionImpact i They are the degree of influence based on the analysis of time and operating point;
[0074] After analyzing the impact, a comprehensive report is generated, including time series analysis charts, operating point characteristic charts, characteristic value segmentation statistics charts, and comparison charts before and after weight adjustment;
[0075] Adjust weights based on the generated comprehensive report, retrain the time series model using the adjusted weights, and evaluate the model's performance through cross-validation to ensure that the model can more accurately reflect the actual working conditions.
[0076] Combining outlier detection, time series analysis, and operating condition analysis, the model can pay more attention to those abnormal but important data points by adjusting the weights, while taking into account the dynamic changes of the time series and the data characteristics under specific operating conditions. This weight adjustment strategy helps to improve the robustness and prediction accuracy of the model, especially when dealing with complex and nonlinear industrial data.
[0077] Spectral clustering is used to analyze time trends. The specific steps are as follows:
[0078] B1. Construct the time series matrix T. The time series matrix T is expressed as:
[0079]
[0080] Among them, t ij represents the characteristic value at the i-th time point and the j-th operating condition, m represents the number of time points, and n represents the number of operating conditions;
[0081] B2. Normalize the time series matrix T. Spectral clustering also includes constructing a similarity matrix W, calculating a degree matrix D, and a Laplace matrix L;
[0082] The similarity matrix W reflects the similarity between nodes;
[0083] The diagonal elements D ii is equal to the sum of all edge weights of the i-th node, where the diagonal element D ii for:
[0084]
[0085] The standardized matrix is clustered using the spectral clustering algorithm to identify trend patterns, and the Plath matrix L is calculated from the degree matrix and the similarity matrix: L = DW;
[0086] The goal of spectral clustering is to minimize the function y on the Laplacian matrix L T Ly, where y is an indicator vector used to indicate which cluster each node belongs to. The optimal clustering result y is expressed as:
[0087] y=arg min y y TLy (4);
[0088] Among them, y is the optimal clustering result;
[0089] B3. Solve for the eigenvector of L and select the eigenvectors corresponding to the first k smallest non-zero eigenvalues for clustering, where k is the expected number of clusters.
[0090] Spectral clustering is a clustering method based on graph theory, which is particularly suitable for processing nonlinearly distributed data sets. The core idea is to regard data points as nodes in a graph, and the similarities (or distances) between nodes constitute the edges of the graph. Spectral clustering achieves clustering by constructing a Laplacian matrix (L) and solving the eigenvectors of the matrix to find the best partition of the data.
[0091] The technical effects of spectral clustering in time series analysis are mainly reflected in the following aspects:
[0092] Pattern Recognition: Spectral clustering can identify underlying patterns in data, even if these patterns are nonlinear. This is crucial for understanding the dynamic behavior of complex systems.
[0093] Anomaly Detection: By identifying normal trends and patterns, spectral clustering can help spot anomalous data points that deviate from these patterns, which is useful for early warning of failures.
[0094] Feature extraction: Spectral clustering can extract key features from raw data. These features may reveal the intrinsic structure of the data set and are very helpful for subsequent predictive modeling.
[0095] In the preferred solution, in the frequency trend analysis, the sliding window technology is used to calculate the frequency of the characteristic value interval. The specific method is:
[0096] Frequency trend analysis is an important part of data analysis, especially when dealing with time series data. It can help us understand the distribution of a certain feature value in different time windows and its trend over time. By dividing the continuous intervals and counting the frequency of the feature values in each interval, an amplitude-frequency histogram can be generated to identify the distribution pattern and anomalies of the feature values.
[0097] Sliding window technology is a commonly used method in frequency trend analysis. It moves a fixed-size window on the time series data and calculates the statistical characteristics of the data in the window in sequence, such as mean, variance, frequency distribution, etc. This method can capture the local changes of data over time and is particularly effective in identifying short-term trends and abnormal patterns.
[0098] The Chi-squared test is a statistical hypothesis testing method, which is mainly used to compare the significant difference between the observed frequency and the expected frequency. In frequency trend analysis, the Chi-squared test can be used to determine whether the distribution of characteristic values in different time windows has changed significantly, thereby identifying the moment of abnormal frequency or trend change.
[0099] C1. Define the sliding window size and step size, slide the window on the time series, and calculate the distribution of feature values in each window;
[0100] C2. Use the chi-square test to assess distribution differences and identify abnormal frequencies.
[0101] Frequency trend analysis combines sliding window technology and chi-square test to understand the distribution of eigenvalues over time and detect anomalies in the data in a timely manner.
[0102] By carefully adjusting the size and step size of the sliding window and reasonably setting the eigenvalue interval, the dynamic characteristics of the data can be captured more accurately and valuable information can be provided to decision makers.
[0103] In the preferred solution, in the correlation analysis, mutual information is used to calculate the correlation between the characteristic values of different measuring points. The specific method is:
[0104] D1, collects characteristic value time series data from each measurement point for a period of time to calculate probability distribution and mutual information;
[0105] D2. For any two measuring points X and Y, calculate the joint probability distribution of the eigenvalues, discretize the data, map the continuous eigenvalues into a finite interval, then count the number of samples falling on each interval combination, and finally normalize to obtain the joint probability distribution;
[0106] D3. Calculate the marginal probability distributions P(X) and P(Y). The marginal probability distribution reflects the probability distribution of the eigenvalue of a single measuring point, that is, the probability of the eigenvalue of a certain measuring point occurring without considering other measuring points. It is obtained by summing the joint probability distribution along another dimension.
[0107] D4. Apply the mutual information formula to calculate the correlation. Use the mutual information formula to quantify the correlation between the eigenvalues of the measurement points X and Y. The mutual information formula is:
[0108]
[0109] Where I(X;Y) represents the mutual information of X and Y, P(x,y) is the joint probability, and P(x) and P(y) are the marginal probabilities respectively;
[0110] D5, repeat the above D1-D4 process for all measurement points in pairs to obtain a complete mutual information matrix;
[0111] Visualize this matrix as a heat map or network diagram to intuitively show the strength of correlation between the eigenvalues of the measurement points;
[0112] D6. Based on the mutual information matrix, identify which measurement point eigenvalues are highly correlated and which measurement points are relatively independent; high mutual information values indicate direct or indirect causal relationships, while low mutual information values indicate that the measurement point eigenvalues are almost independent.
[0113] Fault diagnosis: By identifying the correlation between feature values, the source of abnormal behavior can be traced to assist in fault location.
[0114] Prediction Model: Selecting highly correlated feature values as input can improve the performance of the prediction model.
[0115] System optimization: Understanding the interactions between measurement points helps design more effective control strategies.
[0116] It can deeply explore the complex correlations between the characteristic values of different measurement points, providing strong support for system analysis and decision-making. Through the above steps, a correlation matrix diagram of the characteristic values of the measurement points is constructed to further gain insight into the internal operating mechanism of the system, laying the foundation for subsequent data-driven research and applications.
[0117] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions described in the claims, including equivalent replacement solutions of the technical features in the technical solutions described in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.
Claims
1. A hydropower equipment fault analysis method based on an industrial Internet platform, characterized by: The analysis method steps include working condition analysis, time trend analysis, frequency trend analysis and correlation analysis; Working condition analysis: based on the characteristic value of the measuring point and the historical data of the working condition, determine the interval between the abnormal data of the measuring point and the working condition; Time trend analysis: according to the characteristic value of the exceeding measurement point and the historical data of the corresponding working conditions, the characteristic value of the exceeding measurement point within the time length is taken according to different time lengths, and a time-characteristic value trend chart is generated to analyze the development trend of the characteristic value of the measurement point; Frequency trend analysis, based on the characteristic value of the measuring point and the data of a certain time length of the corresponding working condition, divides the characteristic value within the time into continuous intervals, counts the frequency of the characteristic value of the measuring point in different characteristic value intervals under all working conditions, generates an amplitude-frequency histogram, and determines the frequency of the characteristic value in a certain amplitude interval. Correlation analysis: perform correlation analysis on the eigenvalue time series data of different measuring points to generate a correlation matrix diagram of the eigenvalues of different measuring points. Based on the correlation results, determine whether there is a correlation between the measuring points and estimate which factor is related to the cause of the exceeding of the standard.
2. According to claim 1, a method for analyzing hydropower equipment failure based on an industrial Internet platform is characterized in that: The specific steps of working condition analysis include: A1. Classify the characteristic values of the measuring points and the corresponding working condition information of the measuring points according to the time length, generate a bubble chart of active power-measuring point characteristic value-time period, set the bubble color according to the time length, and analyze and determine the active power and time interval when the measuring point is abnormal; A2. Classify the operating point and the characteristic value of the measuring point at the operating point, generate a bubble chart of active power-water head-measurement point characteristic value, set the bubble color according to the characteristic value, and analyze and determine the interval of water head and active power when the measuring point is abnormal; A3. For the points where the characteristic value data of the measuring points exceed the standard, the points are segmented according to their characteristic values, and a bubble chart of active power-water head-measuring point characteristic value is generated to further determine the operating condition range where the data exceeds the standard.
3. According to claim 2, a method for analyzing hydropower equipment failure based on an industrial Internet platform is characterized in that: In the working condition analysis, steps A1-A3 are comprehensively analyzed, and the analysis results are fed back to the adaptive weight time series modeling. The weight is dynamically adjusted according to the correlation strength between the characteristic value and the working condition. The final adjusted weight is: "w" i =αw′ i +(1-a)(b·TimeImpact i +(1-β)·ConditionImpact i ) (1); Among them, w i " is the final adjusted weight, w′ is the weight adjusted based on the degree of abnormality, α is the abnormality weight coefficient, β is the time impact coefficient, TimeImpact i and ConditionImpact i They are the degree of influence based on the analysis of time and operating point; After analyzing the impact, a comprehensive report is generated, including time series analysis charts, operating point characteristic charts, characteristic value segmentation statistics charts, and comparison charts before and after weight adjustment; Adjust weights based on the generated comprehensive report, retrain the time series model using the adjusted weights, and evaluate the model's performance through cross-validation to ensure that the model can more accurately reflect the actual working conditions.
4. According to claim 1, a method for analyzing hydropower equipment failure based on an industrial Internet platform is characterized in that: Spectral clustering is used to analyze time trends. The specific steps are as follows: B1. Construct a time series matrix T. The time series matrix T is expressed as: Among them, t ij represents the characteristic value at the i-th time point and the j-th operating condition, m represents the number of time points, and n represents the number of operating conditions; B2. Normalize the time series matrix T. Spectral clustering also includes constructing a similarity matrix W, calculating a degree matrix D, and a Laplace matrix L; The similarity matrix W reflects the similarity between nodes; The diagonal elements D ii is equal to the sum of all edge weights of the i-th node, where the diagonal element D ii for: The standardized matrix is clustered using the spectral clustering algorithm to identify trend patterns, and the Plath matrix L is calculated from the degree matrix and the similarity matrix: L = DW; The goal of spectral clustering is to minimize the function y on the Laplacian matrix L T Ly, where y is an indicator vector used to indicate which cluster each node belongs to. The optimal clustering result y is expressed as: y=argmin y the T Ly(4); Among them, y is the optimal clustering result; B3. Solve for the eigenvector of L and select the eigenvectors corresponding to the first k smallest non-zero eigenvalues for clustering, where k is the expected number of clusters.
5. According to claim 1, a method for analyzing hydropower equipment failure based on an industrial Internet platform is characterized in that: In frequency trend analysis, the sliding window technique is used to calculate the frequency of the characteristic value interval. The specific method is: C1. Define the sliding window size and step size, slide the window on the time series, and calculate the distribution of feature values in each window; C2. Use the chi-square test to assess distribution differences and identify abnormal frequencies.
6. According to the method for analyzing hydropower equipment failure based on the industrial Internet platform of claim 1, it is characterized by: In the correlation analysis, mutual information is used to calculate the correlation between the characteristic values of different measurement points. The specific method is: D1, collects characteristic value time series data from each measurement point for a period of time to calculate probability distribution and mutual information; D2. For any two measuring points X and Y, calculate the joint probability distribution of the eigenvalues, discretize the data, map the continuous eigenvalues into a finite interval, then count the number of samples falling on each interval combination, and finally normalize to obtain the joint probability distribution; D3. Calculate the marginal probability distributions P(X) and P(Y). The marginal probability distribution reflects the probability distribution of the eigenvalue of a single measuring point, that is, the probability of the eigenvalue of a certain measuring point occurring without considering other measuring points. It is obtained by summing the joint probability distribution along another dimension. D4. Apply the mutual information formula to calculate the correlation. Use the mutual information formula to quantify the correlation between the eigenvalues of the measurement points X and Y. The mutual information formula is: Where I(X;Y) represents the mutual information of X and Y, P(x,y) is the joint probability, and P(x) and P(y) are the marginal probabilities respectively; D5, repeat the above D1-D4 process for all measurement points in pairs to obtain a complete mutual information matrix; Visualize this matrix as a heat map or network diagram to intuitively show the strength of correlation between the eigenvalues of the measurement points; D6. Based on the mutual information matrix, identify which measurement point eigenvalues are highly correlated and which measurement points are relatively independent; high mutual information values indicate direct or indirect causal relationships, while low mutual information values indicate that the measurement point eigenvalues are almost independent.
Citation Information
Cited By
Hydroelectric equipment multi-modal data analysis early warning method
CN120995073A
